[Ksummit-2008-discuss] A suggestion for Linux 3.0

[Jean Delvare]

Hi Mauro,

On Sat, 30 Aug 2008 16:52:33 -0300, Mauro Carvalho Chehab wrote:
> On Sat, 30 Aug 2008 18:31:44 +0200
> Jean Delvare <khali at linux-fr.org> wrote:
> 
> > Hi Alan,
> > 
> > On Sat, 30 Aug 2008 15:55:31 +0100, Alan Cox wrote:
> > > Having spent ages wading through old broken ISA drivers some of which
> > > clearly have no users as they've not worked for years I want throw in
> > > another proposal for the kernel summit
> > > 
> > > At some point soon we add all the old legacy ISA drivers (barring the odd
> > > ones that turn up in embedded chipsets on LPC bus) into the
> > > feature-removal list and declare an 'ISA death' flag day which we brand
> > > 2.8 or 3.0 or something so everyone knows that we are having a single
> > > clean 'throw out' of old junk.
> > 
> > I don't get the idea. If some drivers are old and unused and we want to
> > delete them, let's just delete them. No need to wait for 2.8. But I
> > fail to see why we should remove all ISA drivers because some ISA
> > drivers are broken. I still use at least one ISA driver (3c509). Do not
> > forget that one strength of Linux is that it runs fine on old hardware.
> > I would certainly like it to stay that way.
> 
> I agree with Alan's proposal.
> 
> Kernel still runs on old hardware, but this doesn't mean that you'll be able to
> actually run any current distro on it. I used to have one of the latest desktop
> from ISA days (1998? hmm.. maybe older) with a hybrid ISA/PCI bus, a pentium
> pro and 80M SRAM (originally, I bought it with 16M) and 1.2 Gb hard drive. 
> 
> The the last time I tried to use that machine (a few years ago), I couldn't find
> any distro with a newer kernel 2.6 that performed fine (and capable on
> installing on its disk), even with a light X11 running on it. For it to work
> with some performance, I needed to use a 2.4 based distro (those distros
> "ready" for 2.6, launched during late 2.5 kernels). That distro were able to
> install on my disk, and allowed me to have a 2.6 kernel on it. It took an entire
> night to compile). After the job, I was almost without free space.
> 
> So, except for us to be proud of that, I can't see any practical sense on
> keeping support for hardware with 10+ years old.

I have a DFI CB60-V3 motherboard here, with a Celeron 400, 96 MB of RAM
and 20 GB of hard disk space. This machine runs perfectly fine under
Slackware 11.0 with a 2.6.26 kernel, and it has 3 ISA slots. Two of
which have a 3Com 3C509 network adapter plugged in. Sure, I'm not
running X11 on it, but it can still serve as a file server or gateway.
The board was produced in 1999, so it's not 10 years old yet.

I have another board here (Asus TX97-E) which also has ISA slots. It
has a Pentium 166 MMX on it and still runs OK. 11 years old.

I'm not saying that there are that many users out there still running
Linux on that kind of hardware. But it happens for sure, and if you are
going to massively drop support for such old but popular hardware, you
should expect loud complaints from a number users.

> So, I really can't see much sense on keep supporting ISA (I would also add MCA)
> bus. IMO, we should put a date for its removal from kernel. If some user still
> needs to use it, it may still run with an older kernel.

I'd rather suggest an equivalent of the EXPERIMENTAL dependency but the
other way around. Say, LEGACY. "Depends on LEGACY" would mean that the
hardware is old enough that we may consider deprecating it, and that in
general we can no longer guarantee it works. Probably commercial
distribution would disable that option to make their kernels smaller
and limit the number of drivers they have to support.

Note that I am not too frightened by Alan's proposal anyway. If we
remove something and there are enough complaints about that move, we
can simply revert the removal. It's never set in the stone.

And I also can't disagree that running 2.4 kernels on such old hardware
is probably better anyway. But it gets hard to find distributions based
on 2.4 kernels and still being maintained. There's Slackware and
probably that's it.

> > > It would also be a chance to throw out a whole pile of other "legacy"
> > > things like ipt_tos, bzImage symlinks, ancient SCTP options, ancient
> > > lmsensor support, V4L1 only driver stuff etc.
> 
> In the case of V4L1, I don't think it would be a good idea just to remove all
> V4L1 only drivers. Legacy V4L stuff may be broken into a few categories:
> 
> 	- Legacy ISA radio - all of them are currently using V4L2 API. We may
> remove it together with ISA removal. Otherwise, I don't see any practical
> reason for their removal, since they just works;
> 
> 	- V4L1 only Zoran driver - AFAIK, there are not-so-old PCI boards with
> those chips. It seems valuable to make they work with V4L2. There's one developer
> working on it, but I don't think this is his top development priority;

The zoran driver is still in wide use, you just can't remove it. If it
needs to be converted, let's convert it.

> 	- a few V4L1 webcam drivers. We need to check case by case. There are
> still some cameras that may be valuable to migrate its support to another
> driver, like gspca. One developer is interested on doing this job;
> 
> 	- legacy PCI ID's on drivers that are still used by newer hardware.
> Bttv is the life example of such hardware: still today companies are selling
> bttv devices. The driver has about 10 years old (I still have an ISA bttv board
> somewhere). I don't see any practical reason for removing support for those devices.

Thanks,
-- 
Jean Delvare